45-EFT.WP.Data.Pipeline v1.0 | Chapter 11 Versioning, Provenance & Lineage

Home ／ Docs-Technical WhitePaper ／ 45-EFT.WP.Data.Pipeline v1.0

Chapter 11 Versioning, Provenance & Lineage

I. Chapter Purpose & Scope

II. Terminology & Dependencies

• Terms: semver, artifact, digest/sha256, lineage.graph, provenance, repro (reproducibility), replay, dual-write, compat_mode.
• Dependencies: contracts/exports (Core.DataSpec v1.0); units/dimensions (Core.Metrology v1.0); splits/coverage & quality (DatasetCards v1.0); evaluation/feature & I/O assumptions (ModelCards v1.0).
• Math & symbols: wrap inline symbols (e.g., QPS, T_inf, ρ) in backticks; any division/integral/composite operator must use parentheses and—if path quantities are involved—declare gamma(ell) and d ell; no Chinese in formulas/symbols/definitions.

III. Fields & Structure (Normative)

versioning:

scheme: "semver" # vMAJOR.MINOR.PATCH

stability_line: "v1.*"

compat_mode: "forward|backward|both|break"

notice:

type: "release|correction|withdrawal"

summary: "<text>"

date: "<YYYY-MM-DD>"

provenance:

sources: ["<uri-or-ref>", "..."] # upstream references (reference-only)

transforms: ["<stage-name>@vX.Y", "..."]

environment:

containers: ["<image@digest>", "..."]

deps_lock: "locks/deps.lock.yaml"

seeds: {global: 1701}

lineage:

graph:

nodes:

- {id:"src.s3.pull", kind:"stage", version:"v1.0"}

- {id:"schema.check", kind:"stage", version:"v1.2"}

- {id:"feat.map", kind:"stage", version:"v1.1"}

- {id:"train_pkg", kind:"artifact", digest:"sha256:..."}

edges:

- {from:"src.s3.pull", to:"schema.check"}

- {from:"schema.check", to:"feat.map"}

- {from:"feat.map", to:"train_pkg"}

replay:

enabled: true

inputs_lock: "locks/inputs.manifest.json" # source list + offsets/watermarks

policy: "strict|lenient"

artifacts:

- {path:"pipeline.yaml", sha256:"<hex>"}

- {path:"locks/inputs.manifest.json", sha256:"<hex>"}

- {path:"locks/deps.lock.yaml", sha256:"<hex>"}

- {path:"outputs/train_pkg.tgz", sha256:"<hex>"}

IV. Versioning Strategy & Stability Line

• Format: vMAJOR.MINOR.PATCH; MAJOR for breaking changes, MINOR for backward-compatible additions, PATCH for fixes and doc corrections.
• Stability line: public references should target v1.*; evaluation/release materials should pin to a minor.
• Compatibility mode: forward|backward|both|break aligned with Schema/contract compat_mode; break requires shadow comparison and notice.
• Notices: record notices under versioning.notice and in export_manifest.references[].

V. Provenance & Reproducibility

• Sources: provenance.sources[] records source identifiers (reference-only); watermarks/offsets and time ranges are locked in inputs.manifest.json.
• Environment: container images use immutable digests; deps.lock.yaml lists dependency versions and hashes.
• Randomness: fix seeds; for non-deterministic operators, declare library versions and deterministic backends.

VI. Lineage Graph & Replay

• Lineage graph: lineage.graph contains nodes (stages/artifacts) and directed edges; nodes carry versions/hashes; the graph must have no dangling nodes.
• Replay: when replay.enabled=true, constrain input sets and order via inputs_lock; policy:"strict" demands byte-identical results, "lenient" allows bounded non-deterministic drift with a stated tolerance.
• Dual-write window: for breaking migrations, use dual-write, compare diffs, and cut over within thresholds.

VII. Artifact Hashing & Integrity

• Mandatory hashing: all critical artifacts (configs/locks/packages/reports) require sha256; integrity failures are blocking.
• Manifest parity: artifacts[] and export_manifest.artifacts[] must match (paths + hashes); optionally record SIZE/LASTMOD.

VIII. Metrology & Units (SI)

1. Performance & resources: QPS (1/s), T_inf (ms {p50,p95,p99}), ρ (—), net_mbps, size_bytes.
2. Mandatory: metrology:{units:"SI", check_dim:true}; normalize units first before composition/conversion.
3. Path quantities: if lineage covers arrival-time/correction chains, register delta_form, path="gamma(ell)", measure="d ell"; use:
   • T_arr = ( 1 / c_ref ) * ( ∫ n_eff d ell ), or
   • T_arr = ( ∫ ( n_eff / c_ref ) d ell ),
     and pass check_dim.

IX. Machine-Readable Fragment (Drop-in)

versioning:

scheme: "semver"

stability_line: "v1.*"

compat_mode: "both"

notice: {type:"release", summary:"initial stable", date:"2025-09-21"}

provenance:

sources: ["s3://eift-data/raw/2025/09/", "contracts/raw_rows@v1.2"]

transforms: ["schema.check@v1.2", "feat.map@v1.1"]

environment:

containers: ["ghcr.io/eift/pipeline@sha256:abcdef..."]

deps_lock: "locks/deps.lock.yaml"

seeds: {global:1701}

lineage:

graph:

nodes:

- {id:"src.s3.pull", kind:"stage", version:"v1.0"}

- {id:"schema.check", kind:"stage", version:"v1.2"}

- {id:"feat.map", kind:"stage", version:"v1.1"}

- {id:"train_pkg", kind:"artifact", digest:"sha256:1234..."}

edges:

- {from:"src.s3.pull", to:"schema.check"}

- {from:"schema.check", to:"feat.map"}

- {from:"feat.map", to:"train_pkg"}

replay: {enabled:true, inputs_lock:"locks/inputs.manifest.json", policy:"strict"}

artifacts:

- {path:"pipeline.yaml", sha256:"..."}

- {path:"locks/inputs.manifest.json", sha256:"..."}

- {path:"locks/deps.lock.yaml", sha256:"..."}

- {path:"outputs/train_pkg.tgz", sha256:"..."}

X. Lint Rules (Excerpt, Normative)

lint_rules:

- id: VER.SEMVER

when: "$.versioning.scheme"

assert: "value == 'semver' and matches($.pipeline.version, '^v\\d+\\.\\d+(\\.\\d+)?$')"

level: error

- id: VER.COMPAT_ALLOWED

when: "$.versioning.compat_mode"

assert: "value in ['forward','backward','both','break']"

level: error

- id: LIN.GRAPH_CONNECTED

when: "$.lineage.graph"

assert: "graph_is_connected(value) and no_dangling_nodes(value)"

level: error

- id: LIN.REPLAY_INPUTS_LOCK

when: "$.lineage.replay.enabled"

assert: "value == false or has_key($.lineage.replay.inputs_lock)"

level: error

- id: ART.SHA256_REQUIRED

when: "$.artifacts[*]"

assert: "has_key('sha256') and len(value.sha256) > 0"

level: error

- id: METROLOGY.SI_AND_CHECKDIM

when: "$.metrology"

assert: "units == 'SI' and check_dim == true"

level: error

XI. Export Manifest & Audit Trail

export_manifest:

version: "v1.0"

artifacts:

- {path:"pipeline.yaml", sha256:"..."}

- {path:"locks/inputs.manifest.json", sha256:"..."}

- {path:"locks/deps.lock.yaml", sha256:"..."}

- {path:"lineage/graph.json", sha256:"..."}

- {path:"reports/replay.result.json", sha256:"..."}

references:

- "EFT.WP.Core.DataSpec v1.0:EXPORT"

- "EFT.WP.Core.Metrology v1.0:check_dim"

- "EFT.WP.Data.DatasetCards v1.0:Ch.11"

- "EFT.WP.Data.ModelCards v1.0:Ch.11"

XII. Chapter Compliance Checklist

• Version follows semver; stability_line matches public references; compat_mode explicit; break changes shadow-compared and noticed.
• Provenance complete: sources, watermarks/offsets, environment & dependency locks, random seeds; no duplication of data facts.
• Lineage graph connected with no dangling nodes; when replay is enabled, provide inputs_lock and strict/lenient posture.
• All critical artifacts carry sha256 and are registered in export_manifest; performance/resource metrology uses SI with check_dim=true.
• For path quantities T_arr, delta_form/path/measure registered and validated; frozen splits and evaluation protocol aligned with Dataset/Model Cards.